My Thesis on AI and Software
AI has unexpectedly become really good at coding. The one industry that is most ripe for destruction is Software as a Service, or SaaS. This was so unforeseen as most thought that physical repetitive jobs that do not require a lot of thought would be replaced first. That is, in part, because a lot of factory work had already been automated and the assumption was that as AI is intelligence, the jobs with less intelligence would be replaced first and eventually move to more complicated intellectual tasks.
That has not been the case. Physical tasks require both a physical component that we have not yet been able to develop, and it requires more processing on various inputs; it is more subjective and has a lot more edge cases that we cannot just program.
Regardless of how we got to where we are, AI is pretty good at coding, and my assumption is that it will continue to get better and better at it. The big question is that are we going to continue to see huge leaps and jumps in Large Language Models (LLM)s in particular, or has progress pretty much plateaued.
If we are still going to see leaps and jumps in improvement, then I think SaaS is going to die. Just as the last 25 years we have seen "software eat the world," and a huge shift from the physical world turning digital, we will see the next 25 years or so where software is so easy to create that there will be no need to buy software from external vendors. There is a realistic scenario where instead of buying an ERP system, you can just ask AI to create you one. As it is completed, then you can have it customize it as much as you want. There are some limitations to this, as large companies are entrenched with certain softwares, but generally speaking if the AI is this good, then it shouldn't matter as the AI can cut through that complexity. Additionally, a lot of industries have requirements on their software, which is why the company will have to gauge if it is cheaper to buy the software or to get a SOC audit, or whatever is required, on the software they custom built with AI.
Lets say AI has plateaued, or is close to plateauing. LLMs is still hugely helpful. It has made coding so easy. I have a friend who has "vibe coded" a platform that has actual paying customers. With this, I was asking about the business and asked what language he is coding in. To this he didn't even know. If it is this easy to code, you are going to see a lot of "pop-up startups" with one person (or a few people) running them building software to challenge large incumbents. Because they are small, they will charge a fraction of what the competitors are charging. There will be a price war and SaaS prices will drop. In this scenario, a good wave to ride is to be one of these pop up shops. In the scenario where AI grows leaps and bound, even these pop up shops will get taken out.
Either way, the next 25 years you will see the commoditization of SaaS, and multiples will fall. Companies that are safe are those with a physical product, a licensed product, or a networked product. Everyone else is in danger.